All research
Hedj RESEARCH·MAY 28, 2026· 5 MIN

A Founder at a 20-Person Startup Isn't a Typical Executive (And an LLM Pre-Training Researcher Isn't a Typical Senior IC)

The level bucket collapsed two very different realities into one number. We now apply additional risk adjustments for founders at their own small startups and for in-demand specialists — capped so no profile reads as zero-risk.

scoringmethodologyfoundersspecialty
KEY FINDINGS
  1. A Founder/Co-Founder at their own solo or SMBcompany can't structurally be laid off. Worth a −10 risk-point adjustment on top of the executive level modifier.
  2. A current hands-on specialist in a scarce, AI-protective field (LLM pre-training, alignment research, GPU systems, chip design, applied cryptography, etc.) earns −4 to −10 depending on rarity.
  3. Combined structural reduction is capped at −20 so no profile floats to the clamp ceiling on structural signals alone. The 5 axes still drive the bulk of the reading.

1 · Why the level bucket isn't enough

The first version of HEDJ assigned the same structural protection to every "executive": −8 risk points. That collapses two very different realities into one number.

A hired CTO at a public companycan absolutely be replaced — the board fires execs all the time. Meta's 2024 reductions cut Director-level employees at the same rate as senior ICs[1]. The −8 bonus is right for them.

A Co-Founder at their own 20-person SaaS startup is structurally different. They own equity. They have hiring and firing power. They're not on a layoff list because there's no list — they run the company. Their displacement risk is whether the product survives, not whether they survive at the company. The −8 bonus is too small.

2 · The founder criterion

HEDJ now applies an additional −10 structural reduction (the Founder Protection chip) when both conditions hold:

  • Current/most-recent roletitle contains "Founder", "Co-Founder", or "Founding" (case-insensitive). Past founders don't qualify — they're at a new company now, and the new company's dynamics apply.
  • Company is still small. company_size_estimate must be solo or smb, OR the profile text clearly indicates fewer than ~200 employees / pre-Series-C stage. Founders of public or late-stage companies face board pressure and can be replaced; they don't qualify.

3 · The specialty premium

The second adjustment recognizes scarce, AI-protective specialties where market demand exceeds supply. The agent looks for hands-on current work in these fields and assigns:

  • −4: niche specialty, some scarcity (e.g. systems performance at a hyperscaler)
  • −7: in-demand specialty (e.g. ML infrastructure at FAANG, GPU kernel work)
  • −10: top-of-market scarcity (e.g. frontier-LLM pre-training, alignment research at a frontier lab)

The qualifying specialty must be both (a) currently in unusually short labor supply and(b) protective against AI displacement — i.e. building the AI rather than being augmented by it. A generic "data scientist" doesn't qualify; a researcher fine-tuning foundation models at OpenAI does.

4 · The combined cap

Founder protection + specialty premium together cap at −20. Level modifier and tenure bonus stack on top. The cap exists because we don't want any profile to read as zero risk — the 5 risk axes drive most of the signal, and structural is a floor adjustment, not a complete override.

5 · Worked examples

Allen — Co-Founder & CTO at GoPlus Security (SMB). Founder protection fires (−10). No specialty premium (Web3 security is broad, not bottlenecked). Level executive (−8). Tenure 6-10y exec (+5). Net structural: −13. Months land in WATCH/STABLE band but visibly below clamp ceiling.

OpenAI pre-training researcher, mid-level IC. No founder protection. Specialty premium fires at −10 (frontier-LLM pre-training). Level mid (0). Tenure 3-5y (+3). Net structural: −7. Reads WATCH even when other axes are high.

Hired CTO at a public software co. Title contains CTO but company_size_estimate = enterprise. No founder protection. No specialty premium. Level executive (−8) is the only structural advantage. Reads as a regular exec.

6 · Limitations

6.1 — LLM-graded.Both adjustments rely on the ScoringAgent reading the profile correctly. We instruct it to be conservative and require explicit evidence; it sometimes over-claims specialty for adjacent fields. We'll tighten the prompt as we see more failure modes.

6.2 — Snapshot, not trajectory. Founder protection treats a 20-person startup the same as a 199-person startup. Future versions could scale the protection with company size + stage signal.

6.3 — Specialty list is seed-driven.The agent has a starting list of qualifying specialties but is also allowed to identify novel scarce specialties it sees on the profile. This will drift over time as labor markets shift; we'll need to refresh the seed list quarterly.

References

  1. [1]
    Meta age-discrimination lawsuit (2024)
    Director-level and IC reductions at similar rates; supports the 'hired exec ≠ founder' distinction.
  2. [2]
    HEDJ research · The 10-Year Tenure Trap
    Companion piece on the tenure × level interaction. Structural adjustments stack on top of tenure and level.